Truckload Optimization via Task Planning based on Deep Reinforcement Learning

ABOUT THIS PROJECT

At a glance

In practice, the truckload optimization problem can be highly complex when taking into account multiple aspects including maximizing capacity of the truck container with package stability, and planning efficient overall routes associated with the order of the loaded packages in the truck. Such a task planning problem can be even more complex with limited and dynamic inputs, where we do not have perfect information of the packages to be loaded, or the target may change according to the current situation. This challenge motivates online task replanning with efficient computation. It can be highly challenging to solve such problems with conventional methods utilized in task planning. On the other hand, deep reinforcement learning (DRL) has shown strong potential in solving highly complex task planning problems with better optimality and computational efficiency. In this project, we plan to leverage the capabilities of DRL to solve the complicated truckload optimization problem. The DRL agent will learn from a truckload simulator with container loading and routing, as well as heuristics and verification via established algorithms based on numerical optimization.